Specifically, there is a Darwin Core Occurrence class, but no separate class to explicitly represent the physical evidence for the Occurrence.
I think the proposed term CollectionObject cannot be a solution to the the problem stated above. Item (organisms) can be evidence of an occurrence, without being present in a collection.
Example: A radio-tagged whale. Tagged or geolocated tree in nature (Zoologists note: very few plants move!).
including "PhysicalObject", "Evidence", "PhysicalEvidence", "Sample", "Specimen", "CollectionObject" in the ASC model, "DataSets/DataSet/Units/Unit/..." in ABCD, "Representation" (Baskauf 2010) and "Token" as a preferred alternative to "Representation" in tdwg-content (http://lists.tdwg.org/pipermail/tdwg-content/2010-October/001742.html).
On Physical Object: Shall digitial vouchers be excluded or not?
If the role is the central concept that needs to be modeled, I would propose something like "Voucher". A CollectionObject can then be a kind of voucher.
Gregor